Fireworks breached twice: $17.5B AI firm admits customer datasets, in-training models accessed
ccerrato147 · x · 2026-11-10
Fireworks, a $17.5B AI inference platform with 10,000+ customers including Harvey, Notion and Cursor, has been hacked twice.
- First break-in (found Oct 6): attackers used internal credentials to read customer API keys and tokens stored for evaluation jobs. The October notice claimed no customer models or datasets were touched; a Nov 9 update admits some were.
- Second break-in: on Oct 23 the cloud hosting Fireworks's research machines was breached, and attackers used it to reach Fireworks's dev machines the next day, running their own jobs.
- Exposure: customer billing records (addresses, invoices, balances, credit use), plus datasets and models still in training for a limited set of customers. No card numbers were stored, so none were taken.
- Attribution: Fireworks links the attack to an underground "LLMjacking" group reselling stolen access to AI models; Google's Mandiant has seen victims billed over $100,000 in a single day.
More from Infra
- Nvidia lobbied hard against rules on remote access to restricted chips, officials say — rohanpaul_ai · 2026-11-10
- AI daily: Fable 5.2 tops evals, orchestrates 1000 agents for 8 days to build a prototype OS — jiqizhixin · 2026-11-10
- Engineer calls megakernel hype a psyop: CUDA graphs already express a strict superset — mike64_t · 2026-11-10
- Senate advances energy permitting bill 73-25, requiring data centers to pay more grid costs — firstadopter · 2026-11-10
- vLLM adopts LibTorch Stable ABI: PyTorch upgrades no longer require rebuilds — vllm_project · 2026-11-10
- Big Tech commits $26B to Anthropic IPO, which targets up to $69.3B raise — SumitGup · 2026-11-10